Complementary metal oxide semiconductor (“CMOS”) image sensors are widely used in digital cameras to produce digital images by converting optical signals into electrical signals. In operation, CMOS image sensors convert an optical signal into an electrical signal using a multitude of pixels that each include a photodiode and a read-out circuit. The photodiode generates electric charges using absorbed light, converts the generated electric charges into an analog current, and delivers the analog current to the read-out circuit. The read-out circuit converts the analog signal into a digital signal and outputs the digital signal.
FIG. 1 illustrates one exemplary design of pixel for a CMOS image sensor that includes four transistors (i.e., a “4T pixel”) that is connected to a bitline. As shown, the CMOS image sensor pixel 10 includes a photodiode (“PD”) that provides the photon to electron conversion, while a floating diffusion (“FD”) point provides the electron to voltage conversion. The voltage per electron conversion of the FD is known as conversion gain (“CG”) and is an important parameter for CMOS image sensors. Conversion gain boosts the pixel signal relative to the analog noise, thereby reducing the noise floor, and thereby enabling performance at lower light levels.
For CMOS image sensors, including the 4T pixel design shown in FIG. 1, digital double sampling (“DDS”) is used to remove undesired offsets of measured pixel voltage output. DDS means obtaining a difference (Dsig−Drst) between digital data Drst obtained by converting a first analog signal output by an initialized pixel into digital data, and digital data Dsig obtained by converting a second analog signal received from the pixel that has received an external image signal into digital data, wherein the second analog signal corresponds to the external image signal. Using the pixel design in FIG. 1 as an example, the pixel is reset when the reset transistor (“RST”) and transfer gate (“TG”) are turned on simultaneously, setting both the floating diffusion FD and the photodiode PD to the VDD voltage level. Next, the transfer gate TG is turned off (disconnecting the photodiode PD and floating diffusion FD) and the photodiode PD is left to integrate light.
After integration, the signal measurement occurs. First, the reset transistor RST is turned on and off to reset the floating diffusion FD. Immediately after this, the reset level is sampled from the floating diffusion FD and stored on the column circuit, i.e., bitline 20. Next, the transfer gate TG is turned on and off which allows charge on the photodiode PD to transfer to the floating diffusion (FD). Once the charge transfer is complete, this charge (the photodiode signal level plus the floating diffusion reset level) is measured and stored on bitline 20 as well.
These two stored voltages are then differenced (Dsig−Drst) to determine the photodiode signal level. This design allows for correlated double sampling (“CDS”) operation to occur, as the reset level used to determine the absolute pixel level is now measured before the signal level and the same reset level is referenced throughout the measurement. Using DDS, a pixel array using the 4T pixel design 10, for example, significantly improves the performance of such CMOS image sensors, reducing both read noise and image lag. In addition, the design reduces pixel source follow offsets and the like.
One technical issue with existing CMOS image sensors is that they can be prone to producing artifacts in the resulting image. Artifacts are areas in the image that correspond to areas of the sensor that have been exposed to extremely high light levels and can appear black (e.g., “black sun”) in the image when in fact they should be the brightest objects in the image.
Artifacts are produced by highly oversaturated pixels in the pixel array. Oversaturated pixels are pixels that are exposed to more light energy than the pixel's photodiode can absorb during exposure time. For example, if a CMOS image sensor is operating with DDS and limiters are not applied, charge can quickly build up after reset so that the sampled dark value fills up. As described above, the difference between the reset voltage (i.e., the dark sample) and the signal voltage (i.e., the bright sample) determines the resulting brightness value of the pixel (i.e., the DDS output). When the dark sample increases significantly due to pixel saturation, the DDS output will drop significantly resulting in black sun artifacts.
FIG. 2A illustrates a graph illustrating the DDS output as a function of the sampled dark and bright pixel voltage outputs. As shown, while the pixel initially operates with full functionality (i.e., no saturation), the DDS output is aligned to the sampled and digitized bright value. However, when the sampled dark value reaches a critical value as the pixel begins to saturate, the DDS output begins to decrease even though the bright value remains constant (e.g., at over 4000 DN). As the dark value increases, the DDS output drops to a point where the resulting output is black even though the measured illumination is very bright. FIG. 2B illustrates a computer image showing black sun artifacts in this case.
Some existing designs have attempted to resolve this issue by adding a limiter on the sample black value. For example, the read out circuit of the pixel array may include logic that provides a DDS output of 2{circumflex over ( )}14−1, for example, if the sampled black value is greater than a certain limit value. However, the resulting image is still not clean as the highlights are washed out and randomized due to raw dark values. FIG. 2C illustrates a computer image showing a resulting image of a CMOS image sensor using limiter circuit to remove black sun artifacts.
Yet another existing design has attempted to use a lookup table fader on the dark signal to remove black sun artifacts. Based on the measured dark value output, a lookup table will define a DDS output and as the dark value increases, the DDS output will also increase in a defined manner. FIG. 2D illustrates an example of the lookup table fader. However, as shown, the highlights still appeared washed out (i.e., the hatched pattern cannot be seen for the portion of saturate pixels) as shown in FIG. 2E. Thus, improvements in such imaging technology to reduce artifacts are still needed.